Title:
LEARNING PROGRAM AND LEARNER
Document Type and Number:
WIPO Patent Application WO/2023/175722
Kind Code:
A1
Abstract:
This learning program is for, in a neural network or a dynamical system, performing calculation for updating an estimated value of a weight or a state variable. The learning program performs: a first calculation for obtaining, from a pre-updated weight, a Karman gain by using an ensemble Karman filter method; a second calculation for adding, to the pre-updated weight, a result obtained by multiplying, by the Karman gain, an error between a training signal and an inference result using the pre-updated weight to obtain a post-updated weight in a first bit expression; and a third calculation for performing bit quantization of the post-updated weight represented by the first bit expression and for changing the result obtained therefrom to a second bit expression in which the word length and the length of a decimal part are shorter than those in the first bit expression.
Inventors:
NAKADA KAZUKI (JP)
Application Number:
PCT/JP2022/011629
Publication Date:
September 21, 2023
Filing Date:
March 15, 2022
Export Citation:
Assignee:
TDK CORP (JP)
International Classes:
G06N3/08
Domestic Patent References:
| WO2020262587A1 | 2020-12-30 |
Foreign References:
| JP2019028746A | 2019-02-21 | |||
| US20220044114A1 | 2022-02-10 |
Other References:
JUNGWOOK CHOI, WANG ZHUO, VENKATARAMANI SWAGATH, CHUANG PIERCE I-JEN, SRINIVASAN VIJAYALAKSHMI, GOPALAKRISHNAN KAILASH: "PACT: Parameterized Clipping Activation for Quantized Neural Networks", 17 July 2018 (2018-07-17), XP055603979, Retrieved from the Internet
Attorney, Agent or Firm:
OIKAWA Shu et al. (JP)
Download PDF: